BUG: concat of objects with the same timezone get reset to UTC

[jreback]

http://stackoverflow.com/questions/24830952/changing-timezone-on-pandas-datetimeindex-when-concatenating-dataframes-in-pytho

ix1 = pd.DatetimeIndex(start=pd.Timestamp('20140715', tz='EST5EDT'), end=pd.Timestamp('20140717', tz='EST5EDT'), freq='D', tz='EST5EDT')
ix2 = pd.DatetimeIndex([pd.Timestamp('2014-07-11 00:00:00', tz='EST5EDT'), pd.Timestamp('2014-07-21 00:00:00', tz='EST5EDT')])
df1 = pd.DataFrame(0, index=ix1, columns=['A', 'B'])
df2 = pd.DataFrame(0, index=ix2, columns=['A', 'B'])

I think the because the 2nd index is only length 2 its freq is not inferred causing a reset to UTC.

In [46]: concat([df1,df2])
Out[46]: 
                           A  B
2014-07-15 04:00:00+00:00  0  0
2014-07-16 04:00:00+00:00  0  0
2014-07-17 04:00:00+00:00  0  0
2014-07-11 04:00:00+00:00  0  0
2014-07-21 04:00:00+00:00  0  0

In [47]: concat([df1,df2]).index
Out[47]: 
<class 'pandas.tseries.index.DatetimeIndex'>
[2014-07-15 04:00:00+00:00, ..., 2014-07-21 04:00:00+00:00]
Length: 5, Freq: None, Timezone: UTC

[sinhrks]

This looks difficult problem than I thought. Because these timezones are distinguished because of DST, need_utc_convert turns True. Need to define how the condition to regard 2 timezones are the same.

ix1.tz, ix2.tz
# <DstTzInfo 'EST5EDT' EDT-1 day, 20:00:00 DST>, <DstTzInfo 'EST5EDT' EST-1 day, 19:00:00 STD>
ix1.tz == ix2.tz
# False

https://github.com/pydata/pandas/blob/master/pandas/tseries/index.py#L2061

[jreback]

@sinhrks I think this is a creation issue in the way DatetimeIndex figures out tz's. It is using an incorrect definition (I think) ix2, though not sure why.

[sinhrks]

Ah I see. Actually each creation and inference results in different tz.

ix1.tz
# <DstTzInfo 'EST5EDT' EDT-1 day, 20:00:00 DST>

ix2.tz
# <DstTzInfo 'EST5EDT' EST-1 day, 19:00:00 STD>

ix3 = pd.DatetimeIndex(start='2014-07-15', end='2014-07-17', freq='D', tz='EST5EDT')
ix3.tz
# <DstTzInfo 'EST5EDT' EST-1 day, 19:00:00 STD>

[jreback]

@sinhrks pr for this?

[Safrone]

I notice this bug also happens when using the + operator on two dataframes

[jreback]

you should file a new issue with s reproducible example then